GNSS Urban Localization Enhencement using Dirichlet Process Mixture Modelling
DUFLOS ; VANHEEGHE ; VIANDIER ; MARAIS ; RABAOUI
Type de document
COMMUNICATION PAR AFFICHE (AFF)
Langue
anglais
Auteur
DUFLOS ; VANHEEGHE ; VIANDIER ; MARAIS ; RABAOUI
Résumé / Abstract
Precise GNSS localization in urban environment is a key point for the development of a new generation of applications based on systems like GPS or Galileo. Whatever the complexity of the systems, measurement noise linked to multipath cannot be taken into account a priori and must be processed locally. We have shown that this noise can be modelled by a mixture of gaussian. Classical receiver based on least squares estimation or Extended kalman Filter are therefore not designed to account for such a type of error. A mixture modelling must then be considered along with a new generation of estimation algorithm. It is however difficult to estimate a priori the right mixture component number. Dirichlet Process Mixture (DPM) is thus a interesting way to manage this problem. We show in this paper that DPM modelling along with bayesian infrence can improve in a significant way the localization performances; it has been proved on both simulated and real data.
